NVIDIA needs high-bandwidth advanced memory chips, and a lot of them, since the astronomical demand for its AI semiconductors is only getting higher. Samsung has launched some powerful products to support that demand, and its focus on this segment of the memory semiconductor market might make it only supplier of HBM3E modules to NVIDIA this year.
Samsung is far from the only chipmaker to produce high-bandwidth memory modules. Other companies in South Korea, including SK Hynix, are in the running as well to secure these lucrative orders, but it appears that Samsung's got a good thing going with NVIDIA.
Samsung might win all orders from NVIDIA for HBM3E memory chips
Samsung has outpaced its competition, including Micron and SK Hynix, by developing the 12-layer HBM3E modules. This might make it possible for the company to leapfrog the two rivals and win all orders from NVIDIA.
Micron announced the mass production of its 8-layer HBM3E last month while Samsung has come out with a 36GB 12-layer module. Despite having a higher layer count, Samsung's module is the same height as the 8-layer HBM3E. Its performance and capacity is 50% higher compared to that module.
Becoming NVIDIA's exclusive supplier for memory products that are in extremely high demand will help improve Samsung's profits from the semiconductor division that have been struggling for the past couple of years. The company is expected to start supplying HBM3E modules to NVIDIA by September 2024, but Samsung isn't confirming any specifics, saying that it doesn't disclose customer information.
NVIDIA seems to be interested in a close collaboration with Samsung as well. It recently sung praises of the company's HBM3E technology. NVIDIA CEO Jensen Huang also paid a visit to Samsung's booth at GTC 2024 and personally signed a display showcasing Samsung HBM3E, calling the advanced memory modules “Jensen approved.”
